CMSManhattan/JiRackUltra_7b

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 1, 2026License:mitArchitecture:Transformer0.0K Open Weights Featherless Exclusive Cold

CMSManhattan/JiRackUltra_7b is a 7.6 billion parameter model built on a DeepSeek R1 -7B architecture, refactored with BitNet features for efficient CPU inference and an updated tokenizer. It supports native ternary (BitNet-style) operations and includes specialized tags for Routing, Media, Vision, Sound, Tool call, and Robotics. This model is optimized for high-quality CPU inference, particularly for agentic and instruct models requiring advanced tool calling capabilities in resource-constrained environments.

Loading preview...

JiRack Ultra 7B: CPU-Optimized Ternary Model

JiRack Ultra 7B is a 7.6 billion parameter model developed by CMSManhattan, built upon a DeepSeek R1 -7B architecture. It is specifically engineered for fast and efficient CPU inference, incorporating BitNet features for native ternary support and an updated tokenizer. This model is designed to be cloud-ready, offering cost savings on cloud infrastructure and serving as an expert model in RAG deployments.

Key Capabilities

  • CPU-Optimized Performance: Leverages BitNet-style ternary architecture for high-quality CPU inference, supporting TQ_2 on Llama.cpp and Ollama via Quantization-Aware Training (QAT).
  • Advanced Tokenization: Features an updated tokenizer with specialized tags for Routing, Media, Vision, Sound, Tool call, and Robotics, enhancing domain-specific applications.
  • Efficient Resource Use: Designed for efficiency, making it suitable for deployments where cost and resource consumption are critical, such as edge devices or cloud environments.
  • Tool Calling Expertise: Adaptable for agentic or instruct models, utilizing its unique tokenizer to enable high-quality tool calling even on smaller models, functioning as a domain-specific tool expert.
  • GGUF Quantizations: Provides ready-to-run GGUF quantizations (Q2_K, Q3_K_M, Q4_K_M) for flexible deployment across various memory constraints.

Good For

  • Cost-Effective Cloud Deployments: Ideal for users looking to reduce cloud infrastructure costs through efficient CPU inference.
  • Agentic and Tool-Calling Applications: Excels in scenarios requiring advanced tool integration, robotics control, routing, and multimedia processing due to its specialized tokenizer.
  • Edge and Resource-Constrained Environments: Its CPU optimization and efficient quantizations make it suitable for deployment on devices with limited computational resources.
  • Custom QAT for Specific Tasks: Offers services for tailored ternary compression via QAT from user datasets, optimizing performance for specific tasks and avoiding catastrophic forgetting.
  • Enterprise Integration: Provides support for Spring Boot AI and GoEx AI tool call libraries, facilitating integration into enterprise systems.

Finetunes
1 models